I was happy to come into this thread and see I was not the first person for whom that quote came to mind. The dial-up Internet comparison implicitly argues for a particular outcome of current AI as a technology, but doesn't actually support that argument.

There's another presumably unintended aspect of the comparison that seems worth considering. The Internet in 2025 is certainly vastly more successful and impactful than the Internet in the mid-90s. But dial-up itself as a technology for accessing the Internet was as much of a dead-end as Laserdisc was for watching movies at home.

Whether or not AI has a similar trajectory as the Internet is separate from the question of whether the current implementation has an actual future. It seems reasonable to me that in the future we're enjoying the benefits of AI while laughing thinking back to the 2025 approach of just throwing more GPUs at the problem in the same way we look back now and get a chuckle out of the idea of "shotgun modems" as the future.